1 /* 2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. 3 * Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved. 4 * 5 * This copyrighted material is made available to anyone wishing to use, 6 * modify, copy, or redistribute it subject to the terms and conditions 7 * of the GNU General Public License version 2. 8 */ 9 10 #include <linux/bio.h> 11 #include <linux/sched.h> 12 #include <linux/slab.h> 13 #include <linux/spinlock.h> 14 #include <linux/completion.h> 15 #include <linux/buffer_head.h> 16 #include <linux/statfs.h> 17 #include <linux/seq_file.h> 18 #include <linux/mount.h> 19 #include <linux/kthread.h> 20 #include <linux/delay.h> 21 #include <linux/gfs2_ondisk.h> 22 #include <linux/crc32.h> 23 #include <linux/time.h> 24 25 #include "gfs2.h" 26 #include "incore.h" 27 #include "bmap.h" 28 #include "dir.h" 29 #include "glock.h" 30 #include "glops.h" 31 #include "inode.h" 32 #include "log.h" 33 #include "meta_io.h" 34 #include "quota.h" 35 #include "recovery.h" 36 #include "rgrp.h" 37 #include "super.h" 38 #include "trans.h" 39 #include "util.h" 40 #include "sys.h" 41 #include "xattr.h" 42 43 #define args_neq(a1, a2, x) ((a1)->ar_##x != (a2)->ar_##x) 44 45 enum { 46 Opt_lockproto, 47 Opt_locktable, 48 Opt_hostdata, 49 Opt_spectator, 50 Opt_ignore_local_fs, 51 Opt_localflocks, 52 Opt_localcaching, 53 Opt_debug, 54 Opt_nodebug, 55 Opt_upgrade, 56 Opt_acl, 57 Opt_noacl, 58 Opt_quota_off, 59 Opt_quota_account, 60 Opt_quota_on, 61 Opt_quota, 62 Opt_noquota, 63 Opt_suiddir, 64 Opt_nosuiddir, 65 Opt_data_writeback, 66 Opt_data_ordered, 67 Opt_meta, 68 Opt_discard, 69 Opt_nodiscard, 70 Opt_commit, 71 Opt_err_withdraw, 72 Opt_err_panic, 73 Opt_error, 74 }; 75 76 static const match_table_t tokens = { 77 {Opt_lockproto, "lockproto=%s"}, 78 {Opt_locktable, "locktable=%s"}, 79 {Opt_hostdata, "hostdata=%s"}, 80 {Opt_spectator, "spectator"}, 81 {Opt_ignore_local_fs, "ignore_local_fs"}, 82 {Opt_localflocks, "localflocks"}, 83 {Opt_localcaching, "localcaching"}, 84 {Opt_debug, "debug"}, 85 {Opt_nodebug, "nodebug"}, 86 {Opt_upgrade, "upgrade"}, 87 {Opt_acl, "acl"}, 88 {Opt_noacl, "noacl"}, 89 {Opt_quota_off, "quota=off"}, 90 {Opt_quota_account, "quota=account"}, 91 {Opt_quota_on, "quota=on"}, 92 {Opt_quota, "quota"}, 93 {Opt_noquota, "noquota"}, 94 {Opt_suiddir, "suiddir"}, 95 {Opt_nosuiddir, "nosuiddir"}, 96 {Opt_data_writeback, "data=writeback"}, 97 {Opt_data_ordered, "data=ordered"}, 98 {Opt_meta, "meta"}, 99 {Opt_discard, "discard"}, 100 {Opt_nodiscard, "nodiscard"}, 101 {Opt_commit, "commit=%d"}, 102 {Opt_err_withdraw, "errors=withdraw"}, 103 {Opt_err_panic, "errors=panic"}, 104 {Opt_error, NULL} 105 }; 106 107 /** 108 * gfs2_mount_args - Parse mount options 109 * @args: The structure into which the parsed options will be written 110 * @options: The options to parse 111 * 112 * Return: errno 113 */ 114 115 int gfs2_mount_args(struct gfs2_args *args, char *options) 116 { 117 char *o; 118 int token; 119 substring_t tmp[MAX_OPT_ARGS]; 120 int rv; 121 122 /* Split the options into tokens with the "," character and 123 process them */ 124 125 while (1) { 126 o = strsep(&options, ","); 127 if (o == NULL) 128 break; 129 if (*o == '\0') 130 continue; 131 132 token = match_token(o, tokens, tmp); 133 switch (token) { 134 case Opt_lockproto: 135 match_strlcpy(args->ar_lockproto, &tmp[0], 136 GFS2_LOCKNAME_LEN); 137 break; 138 case Opt_locktable: 139 match_strlcpy(args->ar_locktable, &tmp[0], 140 GFS2_LOCKNAME_LEN); 141 break; 142 case Opt_hostdata: 143 match_strlcpy(args->ar_hostdata, &tmp[0], 144 GFS2_LOCKNAME_LEN); 145 break; 146 case Opt_spectator: 147 args->ar_spectator = 1; 148 break; 149 case Opt_ignore_local_fs: 150 args->ar_ignore_local_fs = 1; 151 break; 152 case Opt_localflocks: 153 args->ar_localflocks = 1; 154 break; 155 case Opt_localcaching: 156 args->ar_localcaching = 1; 157 break; 158 case Opt_debug: 159 if (args->ar_errors == GFS2_ERRORS_PANIC) { 160 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic " 161 "are mutually exclusive.\n"); 162 return -EINVAL; 163 } 164 args->ar_debug = 1; 165 break; 166 case Opt_nodebug: 167 args->ar_debug = 0; 168 break; 169 case Opt_upgrade: 170 args->ar_upgrade = 1; 171 break; 172 case Opt_acl: 173 args->ar_posix_acl = 1; 174 break; 175 case Opt_noacl: 176 args->ar_posix_acl = 0; 177 break; 178 case Opt_quota_off: 179 case Opt_noquota: 180 args->ar_quota = GFS2_QUOTA_OFF; 181 break; 182 case Opt_quota_account: 183 args->ar_quota = GFS2_QUOTA_ACCOUNT; 184 break; 185 case Opt_quota_on: 186 case Opt_quota: 187 args->ar_quota = GFS2_QUOTA_ON; 188 break; 189 case Opt_suiddir: 190 args->ar_suiddir = 1; 191 break; 192 case Opt_nosuiddir: 193 args->ar_suiddir = 0; 194 break; 195 case Opt_data_writeback: 196 args->ar_data = GFS2_DATA_WRITEBACK; 197 break; 198 case Opt_data_ordered: 199 args->ar_data = GFS2_DATA_ORDERED; 200 break; 201 case Opt_meta: 202 args->ar_meta = 1; 203 break; 204 case Opt_discard: 205 args->ar_discard = 1; 206 break; 207 case Opt_nodiscard: 208 args->ar_discard = 0; 209 break; 210 case Opt_commit: 211 rv = match_int(&tmp[0], &args->ar_commit); 212 if (rv || args->ar_commit <= 0) { 213 printk(KERN_WARNING "GFS2: commit mount option requires a positive numeric argument\n"); 214 return rv ? rv : -EINVAL; 215 } 216 break; 217 case Opt_err_withdraw: 218 args->ar_errors = GFS2_ERRORS_WITHDRAW; 219 break; 220 case Opt_err_panic: 221 if (args->ar_debug) { 222 printk(KERN_WARNING "GFS2: -o debug and -o errors=panic " 223 "are mutually exclusive.\n"); 224 return -EINVAL; 225 } 226 args->ar_errors = GFS2_ERRORS_PANIC; 227 break; 228 case Opt_error: 229 default: 230 printk(KERN_WARNING "GFS2: invalid mount option: %s\n", o); 231 return -EINVAL; 232 } 233 } 234 235 return 0; 236 } 237 238 /** 239 * gfs2_jindex_free - Clear all the journal index information 240 * @sdp: The GFS2 superblock 241 * 242 */ 243 244 void gfs2_jindex_free(struct gfs2_sbd *sdp) 245 { 246 struct list_head list, *head; 247 struct gfs2_jdesc *jd; 248 struct gfs2_journal_extent *jext; 249 250 spin_lock(&sdp->sd_jindex_spin); 251 list_add(&list, &sdp->sd_jindex_list); 252 list_del_init(&sdp->sd_jindex_list); 253 sdp->sd_journals = 0; 254 spin_unlock(&sdp->sd_jindex_spin); 255 256 while (!list_empty(&list)) { 257 jd = list_entry(list.next, struct gfs2_jdesc, jd_list); 258 head = &jd->extent_list; 259 while (!list_empty(head)) { 260 jext = list_entry(head->next, 261 struct gfs2_journal_extent, 262 extent_list); 263 list_del(&jext->extent_list); 264 kfree(jext); 265 } 266 list_del(&jd->jd_list); 267 iput(jd->jd_inode); 268 kfree(jd); 269 } 270 } 271 272 static struct gfs2_jdesc *jdesc_find_i(struct list_head *head, unsigned int jid) 273 { 274 struct gfs2_jdesc *jd; 275 int found = 0; 276 277 list_for_each_entry(jd, head, jd_list) { 278 if (jd->jd_jid == jid) { 279 found = 1; 280 break; 281 } 282 } 283 284 if (!found) 285 jd = NULL; 286 287 return jd; 288 } 289 290 struct gfs2_jdesc *gfs2_jdesc_find(struct gfs2_sbd *sdp, unsigned int jid) 291 { 292 struct gfs2_jdesc *jd; 293 294 spin_lock(&sdp->sd_jindex_spin); 295 jd = jdesc_find_i(&sdp->sd_jindex_list, jid); 296 spin_unlock(&sdp->sd_jindex_spin); 297 298 return jd; 299 } 300 301 int gfs2_jdesc_check(struct gfs2_jdesc *jd) 302 { 303 struct gfs2_inode *ip = GFS2_I(jd->jd_inode); 304 struct gfs2_sbd *sdp = GFS2_SB(jd->jd_inode); 305 int ar; 306 int error; 307 308 if (ip->i_disksize < (8 << 20) || ip->i_disksize > (1 << 30) || 309 (ip->i_disksize & (sdp->sd_sb.sb_bsize - 1))) { 310 gfs2_consist_inode(ip); 311 return -EIO; 312 } 313 jd->jd_blocks = ip->i_disksize >> sdp->sd_sb.sb_bsize_shift; 314 315 error = gfs2_write_alloc_required(ip, 0, ip->i_disksize, &ar); 316 if (!error && ar) { 317 gfs2_consist_inode(ip); 318 error = -EIO; 319 } 320 321 return error; 322 } 323 324 /** 325 * gfs2_make_fs_rw - Turn a Read-Only FS into a Read-Write one 326 * @sdp: the filesystem 327 * 328 * Returns: errno 329 */ 330 331 int gfs2_make_fs_rw(struct gfs2_sbd *sdp) 332 { 333 struct gfs2_inode *ip = GFS2_I(sdp->sd_jdesc->jd_inode); 334 struct gfs2_glock *j_gl = ip->i_gl; 335 struct gfs2_holder t_gh; 336 struct gfs2_log_header_host head; 337 int error; 338 339 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, 0, &t_gh); 340 if (error) 341 return error; 342 343 j_gl->gl_ops->go_inval(j_gl, DIO_METADATA); 344 345 error = gfs2_find_jhead(sdp->sd_jdesc, &head); 346 if (error) 347 goto fail; 348 349 if (!(head.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) { 350 gfs2_consist(sdp); 351 error = -EIO; 352 goto fail; 353 } 354 355 /* Initialize some head of the log stuff */ 356 sdp->sd_log_sequence = head.lh_sequence + 1; 357 gfs2_log_pointers_init(sdp, head.lh_blkno); 358 359 error = gfs2_quota_init(sdp); 360 if (error) 361 goto fail; 362 363 set_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); 364 365 gfs2_glock_dq_uninit(&t_gh); 366 367 return 0; 368 369 fail: 370 t_gh.gh_flags |= GL_NOCACHE; 371 gfs2_glock_dq_uninit(&t_gh); 372 373 return error; 374 } 375 376 void gfs2_statfs_change_in(struct gfs2_statfs_change_host *sc, const void *buf) 377 { 378 const struct gfs2_statfs_change *str = buf; 379 380 sc->sc_total = be64_to_cpu(str->sc_total); 381 sc->sc_free = be64_to_cpu(str->sc_free); 382 sc->sc_dinodes = be64_to_cpu(str->sc_dinodes); 383 } 384 385 static void gfs2_statfs_change_out(const struct gfs2_statfs_change_host *sc, void *buf) 386 { 387 struct gfs2_statfs_change *str = buf; 388 389 str->sc_total = cpu_to_be64(sc->sc_total); 390 str->sc_free = cpu_to_be64(sc->sc_free); 391 str->sc_dinodes = cpu_to_be64(sc->sc_dinodes); 392 } 393 394 int gfs2_statfs_init(struct gfs2_sbd *sdp) 395 { 396 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); 397 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; 398 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); 399 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 400 struct buffer_head *m_bh, *l_bh; 401 struct gfs2_holder gh; 402 int error; 403 404 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE, 405 &gh); 406 if (error) 407 return error; 408 409 error = gfs2_meta_inode_buffer(m_ip, &m_bh); 410 if (error) 411 goto out; 412 413 if (sdp->sd_args.ar_spectator) { 414 spin_lock(&sdp->sd_statfs_spin); 415 gfs2_statfs_change_in(m_sc, m_bh->b_data + 416 sizeof(struct gfs2_dinode)); 417 spin_unlock(&sdp->sd_statfs_spin); 418 } else { 419 error = gfs2_meta_inode_buffer(l_ip, &l_bh); 420 if (error) 421 goto out_m_bh; 422 423 spin_lock(&sdp->sd_statfs_spin); 424 gfs2_statfs_change_in(m_sc, m_bh->b_data + 425 sizeof(struct gfs2_dinode)); 426 gfs2_statfs_change_in(l_sc, l_bh->b_data + 427 sizeof(struct gfs2_dinode)); 428 spin_unlock(&sdp->sd_statfs_spin); 429 430 brelse(l_bh); 431 } 432 433 out_m_bh: 434 brelse(m_bh); 435 out: 436 gfs2_glock_dq_uninit(&gh); 437 return 0; 438 } 439 440 void gfs2_statfs_change(struct gfs2_sbd *sdp, s64 total, s64 free, 441 s64 dinodes) 442 { 443 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); 444 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 445 struct buffer_head *l_bh; 446 int error; 447 448 error = gfs2_meta_inode_buffer(l_ip, &l_bh); 449 if (error) 450 return; 451 452 gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1); 453 454 spin_lock(&sdp->sd_statfs_spin); 455 l_sc->sc_total += total; 456 l_sc->sc_free += free; 457 l_sc->sc_dinodes += dinodes; 458 gfs2_statfs_change_out(l_sc, l_bh->b_data + sizeof(struct gfs2_dinode)); 459 spin_unlock(&sdp->sd_statfs_spin); 460 461 brelse(l_bh); 462 } 463 464 void update_statfs(struct gfs2_sbd *sdp, struct buffer_head *m_bh, 465 struct buffer_head *l_bh) 466 { 467 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); 468 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); 469 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; 470 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 471 472 gfs2_trans_add_bh(l_ip->i_gl, l_bh, 1); 473 474 spin_lock(&sdp->sd_statfs_spin); 475 m_sc->sc_total += l_sc->sc_total; 476 m_sc->sc_free += l_sc->sc_free; 477 m_sc->sc_dinodes += l_sc->sc_dinodes; 478 memset(l_sc, 0, sizeof(struct gfs2_statfs_change)); 479 memset(l_bh->b_data + sizeof(struct gfs2_dinode), 480 0, sizeof(struct gfs2_statfs_change)); 481 spin_unlock(&sdp->sd_statfs_spin); 482 483 gfs2_trans_add_bh(m_ip->i_gl, m_bh, 1); 484 gfs2_statfs_change_out(m_sc, m_bh->b_data + sizeof(struct gfs2_dinode)); 485 } 486 487 int gfs2_statfs_sync(struct gfs2_sbd *sdp) 488 { 489 struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); 490 struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); 491 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; 492 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 493 struct gfs2_holder gh; 494 struct buffer_head *m_bh, *l_bh; 495 int error; 496 497 error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, GL_NOCACHE, 498 &gh); 499 if (error) 500 return error; 501 502 error = gfs2_meta_inode_buffer(m_ip, &m_bh); 503 if (error) 504 goto out; 505 506 spin_lock(&sdp->sd_statfs_spin); 507 gfs2_statfs_change_in(m_sc, m_bh->b_data + 508 sizeof(struct gfs2_dinode)); 509 if (!l_sc->sc_total && !l_sc->sc_free && !l_sc->sc_dinodes) { 510 spin_unlock(&sdp->sd_statfs_spin); 511 goto out_bh; 512 } 513 spin_unlock(&sdp->sd_statfs_spin); 514 515 error = gfs2_meta_inode_buffer(l_ip, &l_bh); 516 if (error) 517 goto out_bh; 518 519 error = gfs2_trans_begin(sdp, 2 * RES_DINODE, 0); 520 if (error) 521 goto out_bh2; 522 523 update_statfs(sdp, m_bh, l_bh); 524 525 gfs2_trans_end(sdp); 526 527 out_bh2: 528 brelse(l_bh); 529 out_bh: 530 brelse(m_bh); 531 out: 532 gfs2_glock_dq_uninit(&gh); 533 return error; 534 } 535 536 struct lfcc { 537 struct list_head list; 538 struct gfs2_holder gh; 539 }; 540 541 /** 542 * gfs2_lock_fs_check_clean - Stop all writes to the FS and check that all 543 * journals are clean 544 * @sdp: the file system 545 * @state: the state to put the transaction lock into 546 * @t_gh: the hold on the transaction lock 547 * 548 * Returns: errno 549 */ 550 551 static int gfs2_lock_fs_check_clean(struct gfs2_sbd *sdp, 552 struct gfs2_holder *t_gh) 553 { 554 struct gfs2_inode *ip; 555 struct gfs2_jdesc *jd; 556 struct lfcc *lfcc; 557 LIST_HEAD(list); 558 struct gfs2_log_header_host lh; 559 int error; 560 561 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { 562 lfcc = kmalloc(sizeof(struct lfcc), GFP_KERNEL); 563 if (!lfcc) { 564 error = -ENOMEM; 565 goto out; 566 } 567 ip = GFS2_I(jd->jd_inode); 568 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, 0, &lfcc->gh); 569 if (error) { 570 kfree(lfcc); 571 goto out; 572 } 573 list_add(&lfcc->list, &list); 574 } 575 576 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_DEFERRED, 577 GL_NOCACHE, t_gh); 578 579 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { 580 error = gfs2_jdesc_check(jd); 581 if (error) 582 break; 583 error = gfs2_find_jhead(jd, &lh); 584 if (error) 585 break; 586 if (!(lh.lh_flags & GFS2_LOG_HEAD_UNMOUNT)) { 587 error = -EBUSY; 588 break; 589 } 590 } 591 592 if (error) 593 gfs2_glock_dq_uninit(t_gh); 594 595 out: 596 while (!list_empty(&list)) { 597 lfcc = list_entry(list.next, struct lfcc, list); 598 list_del(&lfcc->list); 599 gfs2_glock_dq_uninit(&lfcc->gh); 600 kfree(lfcc); 601 } 602 return error; 603 } 604 605 /** 606 * gfs2_freeze_fs - freezes the file system 607 * @sdp: the file system 608 * 609 * This function flushes data and meta data for all machines by 610 * aquiring the transaction log exclusively. All journals are 611 * ensured to be in a clean state as well. 612 * 613 * Returns: errno 614 */ 615 616 int gfs2_freeze_fs(struct gfs2_sbd *sdp) 617 { 618 int error = 0; 619 620 mutex_lock(&sdp->sd_freeze_lock); 621 622 if (!sdp->sd_freeze_count++) { 623 error = gfs2_lock_fs_check_clean(sdp, &sdp->sd_freeze_gh); 624 if (error) 625 sdp->sd_freeze_count--; 626 } 627 628 mutex_unlock(&sdp->sd_freeze_lock); 629 630 return error; 631 } 632 633 /** 634 * gfs2_unfreeze_fs - unfreezes the file system 635 * @sdp: the file system 636 * 637 * This function allows the file system to proceed by unlocking 638 * the exclusively held transaction lock. Other GFS2 nodes are 639 * now free to acquire the lock shared and go on with their lives. 640 * 641 */ 642 643 void gfs2_unfreeze_fs(struct gfs2_sbd *sdp) 644 { 645 mutex_lock(&sdp->sd_freeze_lock); 646 647 if (sdp->sd_freeze_count && !--sdp->sd_freeze_count) 648 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh); 649 650 mutex_unlock(&sdp->sd_freeze_lock); 651 } 652 653 654 /** 655 * gfs2_write_inode - Make sure the inode is stable on the disk 656 * @inode: The inode 657 * @sync: synchronous write flag 658 * 659 * Returns: errno 660 */ 661 662 static int gfs2_write_inode(struct inode *inode, int sync) 663 { 664 struct gfs2_inode *ip = GFS2_I(inode); 665 struct gfs2_sbd *sdp = GFS2_SB(inode); 666 struct gfs2_holder gh; 667 struct buffer_head *bh; 668 struct timespec atime; 669 struct gfs2_dinode *di; 670 int ret = 0; 671 672 /* Check this is a "normal" inode, etc */ 673 if (!test_bit(GIF_USER, &ip->i_flags) || 674 (current->flags & PF_MEMALLOC)) 675 return 0; 676 ret = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 677 if (ret) 678 goto do_flush; 679 ret = gfs2_trans_begin(sdp, RES_DINODE, 0); 680 if (ret) 681 goto do_unlock; 682 ret = gfs2_meta_inode_buffer(ip, &bh); 683 if (ret == 0) { 684 di = (struct gfs2_dinode *)bh->b_data; 685 atime.tv_sec = be64_to_cpu(di->di_atime); 686 atime.tv_nsec = be32_to_cpu(di->di_atime_nsec); 687 if (timespec_compare(&inode->i_atime, &atime) > 0) { 688 gfs2_trans_add_bh(ip->i_gl, bh, 1); 689 gfs2_dinode_out(ip, bh->b_data); 690 } 691 brelse(bh); 692 } 693 gfs2_trans_end(sdp); 694 do_unlock: 695 gfs2_glock_dq_uninit(&gh); 696 do_flush: 697 if (sync != 0) 698 gfs2_log_flush(GFS2_SB(inode), ip->i_gl); 699 return ret; 700 } 701 702 /** 703 * gfs2_make_fs_ro - Turn a Read-Write FS into a Read-Only one 704 * @sdp: the filesystem 705 * 706 * Returns: errno 707 */ 708 709 static int gfs2_make_fs_ro(struct gfs2_sbd *sdp) 710 { 711 struct gfs2_holder t_gh; 712 int error; 713 714 flush_workqueue(gfs2_delete_workqueue); 715 gfs2_quota_sync(sdp); 716 gfs2_statfs_sync(sdp); 717 718 error = gfs2_glock_nq_init(sdp->sd_trans_gl, LM_ST_SHARED, GL_NOCACHE, 719 &t_gh); 720 if (error && !test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) 721 return error; 722 723 gfs2_meta_syncfs(sdp); 724 gfs2_log_shutdown(sdp); 725 726 clear_bit(SDF_JOURNAL_LIVE, &sdp->sd_flags); 727 728 if (t_gh.gh_gl) 729 gfs2_glock_dq_uninit(&t_gh); 730 731 gfs2_quota_cleanup(sdp); 732 733 return error; 734 } 735 736 static int gfs2_umount_recovery_wait(void *word) 737 { 738 schedule(); 739 return 0; 740 } 741 742 /** 743 * gfs2_put_super - Unmount the filesystem 744 * @sb: The VFS superblock 745 * 746 */ 747 748 static void gfs2_put_super(struct super_block *sb) 749 { 750 struct gfs2_sbd *sdp = sb->s_fs_info; 751 int error; 752 struct gfs2_jdesc *jd; 753 754 /* Unfreeze the filesystem, if we need to */ 755 756 mutex_lock(&sdp->sd_freeze_lock); 757 if (sdp->sd_freeze_count) 758 gfs2_glock_dq_uninit(&sdp->sd_freeze_gh); 759 mutex_unlock(&sdp->sd_freeze_lock); 760 761 /* No more recovery requests */ 762 set_bit(SDF_NORECOVERY, &sdp->sd_flags); 763 smp_mb(); 764 765 /* Wait on outstanding recovery */ 766 restart: 767 spin_lock(&sdp->sd_jindex_spin); 768 list_for_each_entry(jd, &sdp->sd_jindex_list, jd_list) { 769 if (!test_bit(JDF_RECOVERY, &jd->jd_flags)) 770 continue; 771 spin_unlock(&sdp->sd_jindex_spin); 772 wait_on_bit(&jd->jd_flags, JDF_RECOVERY, 773 gfs2_umount_recovery_wait, TASK_UNINTERRUPTIBLE); 774 goto restart; 775 } 776 spin_unlock(&sdp->sd_jindex_spin); 777 778 kthread_stop(sdp->sd_quotad_process); 779 kthread_stop(sdp->sd_logd_process); 780 781 if (!(sb->s_flags & MS_RDONLY)) { 782 error = gfs2_make_fs_ro(sdp); 783 if (error) 784 gfs2_io_error(sdp); 785 } 786 /* At this point, we're through modifying the disk */ 787 788 /* Release stuff */ 789 790 iput(sdp->sd_jindex); 791 iput(sdp->sd_statfs_inode); 792 iput(sdp->sd_rindex); 793 iput(sdp->sd_quota_inode); 794 795 gfs2_glock_put(sdp->sd_rename_gl); 796 gfs2_glock_put(sdp->sd_trans_gl); 797 798 if (!sdp->sd_args.ar_spectator) { 799 gfs2_glock_dq_uninit(&sdp->sd_journal_gh); 800 gfs2_glock_dq_uninit(&sdp->sd_jinode_gh); 801 gfs2_glock_dq_uninit(&sdp->sd_sc_gh); 802 gfs2_glock_dq_uninit(&sdp->sd_qc_gh); 803 iput(sdp->sd_sc_inode); 804 iput(sdp->sd_qc_inode); 805 } 806 807 gfs2_glock_dq_uninit(&sdp->sd_live_gh); 808 gfs2_clear_rgrpd(sdp); 809 gfs2_jindex_free(sdp); 810 /* Take apart glock structures and buffer lists */ 811 gfs2_gl_hash_clear(sdp); 812 /* Unmount the locking protocol */ 813 gfs2_lm_unmount(sdp); 814 815 /* At this point, we're through participating in the lockspace */ 816 gfs2_sys_fs_del(sdp); 817 } 818 819 /** 820 * gfs2_sync_fs - sync the filesystem 821 * @sb: the superblock 822 * 823 * Flushes the log to disk. 824 */ 825 826 static int gfs2_sync_fs(struct super_block *sb, int wait) 827 { 828 if (wait && sb->s_fs_info) 829 gfs2_log_flush(sb->s_fs_info, NULL); 830 return 0; 831 } 832 833 /** 834 * gfs2_freeze - prevent further writes to the filesystem 835 * @sb: the VFS structure for the filesystem 836 * 837 */ 838 839 static int gfs2_freeze(struct super_block *sb) 840 { 841 struct gfs2_sbd *sdp = sb->s_fs_info; 842 int error; 843 844 if (test_bit(SDF_SHUTDOWN, &sdp->sd_flags)) 845 return -EINVAL; 846 847 for (;;) { 848 error = gfs2_freeze_fs(sdp); 849 if (!error) 850 break; 851 852 switch (error) { 853 case -EBUSY: 854 fs_err(sdp, "waiting for recovery before freeze\n"); 855 break; 856 857 default: 858 fs_err(sdp, "error freezing FS: %d\n", error); 859 break; 860 } 861 862 fs_err(sdp, "retrying...\n"); 863 msleep(1000); 864 } 865 return 0; 866 } 867 868 /** 869 * gfs2_unfreeze - reallow writes to the filesystem 870 * @sb: the VFS structure for the filesystem 871 * 872 */ 873 874 static int gfs2_unfreeze(struct super_block *sb) 875 { 876 gfs2_unfreeze_fs(sb->s_fs_info); 877 return 0; 878 } 879 880 /** 881 * statfs_fill - fill in the sg for a given RG 882 * @rgd: the RG 883 * @sc: the sc structure 884 * 885 * Returns: 0 on success, -ESTALE if the LVB is invalid 886 */ 887 888 static int statfs_slow_fill(struct gfs2_rgrpd *rgd, 889 struct gfs2_statfs_change_host *sc) 890 { 891 gfs2_rgrp_verify(rgd); 892 sc->sc_total += rgd->rd_data; 893 sc->sc_free += rgd->rd_free; 894 sc->sc_dinodes += rgd->rd_dinodes; 895 return 0; 896 } 897 898 /** 899 * gfs2_statfs_slow - Stat a filesystem using asynchronous locking 900 * @sdp: the filesystem 901 * @sc: the sc info that will be returned 902 * 903 * Any error (other than a signal) will cause this routine to fall back 904 * to the synchronous version. 905 * 906 * FIXME: This really shouldn't busy wait like this. 907 * 908 * Returns: errno 909 */ 910 911 static int gfs2_statfs_slow(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc) 912 { 913 struct gfs2_holder ri_gh; 914 struct gfs2_rgrpd *rgd_next; 915 struct gfs2_holder *gha, *gh; 916 unsigned int slots = 64; 917 unsigned int x; 918 int done; 919 int error = 0, err; 920 921 memset(sc, 0, sizeof(struct gfs2_statfs_change_host)); 922 gha = kcalloc(slots, sizeof(struct gfs2_holder), GFP_KERNEL); 923 if (!gha) 924 return -ENOMEM; 925 926 error = gfs2_rindex_hold(sdp, &ri_gh); 927 if (error) 928 goto out; 929 930 rgd_next = gfs2_rgrpd_get_first(sdp); 931 932 for (;;) { 933 done = 1; 934 935 for (x = 0; x < slots; x++) { 936 gh = gha + x; 937 938 if (gh->gh_gl && gfs2_glock_poll(gh)) { 939 err = gfs2_glock_wait(gh); 940 if (err) { 941 gfs2_holder_uninit(gh); 942 error = err; 943 } else { 944 if (!error) 945 error = statfs_slow_fill( 946 gh->gh_gl->gl_object, sc); 947 gfs2_glock_dq_uninit(gh); 948 } 949 } 950 951 if (gh->gh_gl) 952 done = 0; 953 else if (rgd_next && !error) { 954 error = gfs2_glock_nq_init(rgd_next->rd_gl, 955 LM_ST_SHARED, 956 GL_ASYNC, 957 gh); 958 rgd_next = gfs2_rgrpd_get_next(rgd_next); 959 done = 0; 960 } 961 962 if (signal_pending(current)) 963 error = -ERESTARTSYS; 964 } 965 966 if (done) 967 break; 968 969 yield(); 970 } 971 972 gfs2_glock_dq_uninit(&ri_gh); 973 974 out: 975 kfree(gha); 976 return error; 977 } 978 979 /** 980 * gfs2_statfs_i - Do a statfs 981 * @sdp: the filesystem 982 * @sg: the sg structure 983 * 984 * Returns: errno 985 */ 986 987 static int gfs2_statfs_i(struct gfs2_sbd *sdp, struct gfs2_statfs_change_host *sc) 988 { 989 struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; 990 struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; 991 992 spin_lock(&sdp->sd_statfs_spin); 993 994 *sc = *m_sc; 995 sc->sc_total += l_sc->sc_total; 996 sc->sc_free += l_sc->sc_free; 997 sc->sc_dinodes += l_sc->sc_dinodes; 998 999 spin_unlock(&sdp->sd_statfs_spin); 1000 1001 if (sc->sc_free < 0) 1002 sc->sc_free = 0; 1003 if (sc->sc_free > sc->sc_total) 1004 sc->sc_free = sc->sc_total; 1005 if (sc->sc_dinodes < 0) 1006 sc->sc_dinodes = 0; 1007 1008 return 0; 1009 } 1010 1011 /** 1012 * gfs2_statfs - Gather and return stats about the filesystem 1013 * @sb: The superblock 1014 * @statfsbuf: The buffer 1015 * 1016 * Returns: 0 on success or error code 1017 */ 1018 1019 static int gfs2_statfs(struct dentry *dentry, struct kstatfs *buf) 1020 { 1021 struct super_block *sb = dentry->d_inode->i_sb; 1022 struct gfs2_sbd *sdp = sb->s_fs_info; 1023 struct gfs2_statfs_change_host sc; 1024 int error; 1025 1026 if (gfs2_tune_get(sdp, gt_statfs_slow)) 1027 error = gfs2_statfs_slow(sdp, &sc); 1028 else 1029 error = gfs2_statfs_i(sdp, &sc); 1030 1031 if (error) 1032 return error; 1033 1034 buf->f_type = GFS2_MAGIC; 1035 buf->f_bsize = sdp->sd_sb.sb_bsize; 1036 buf->f_blocks = sc.sc_total; 1037 buf->f_bfree = sc.sc_free; 1038 buf->f_bavail = sc.sc_free; 1039 buf->f_files = sc.sc_dinodes + sc.sc_free; 1040 buf->f_ffree = sc.sc_free; 1041 buf->f_namelen = GFS2_FNAMESIZE; 1042 1043 return 0; 1044 } 1045 1046 /** 1047 * gfs2_remount_fs - called when the FS is remounted 1048 * @sb: the filesystem 1049 * @flags: the remount flags 1050 * @data: extra data passed in (not used right now) 1051 * 1052 * Returns: errno 1053 */ 1054 1055 static int gfs2_remount_fs(struct super_block *sb, int *flags, char *data) 1056 { 1057 struct gfs2_sbd *sdp = sb->s_fs_info; 1058 struct gfs2_args args = sdp->sd_args; /* Default to current settings */ 1059 struct gfs2_tune *gt = &sdp->sd_tune; 1060 int error; 1061 1062 spin_lock(>->gt_spin); 1063 args.ar_commit = gt->gt_log_flush_secs; 1064 spin_unlock(>->gt_spin); 1065 error = gfs2_mount_args(&args, data); 1066 if (error) 1067 return error; 1068 1069 /* Not allowed to change locking details */ 1070 if (strcmp(args.ar_lockproto, sdp->sd_args.ar_lockproto) || 1071 strcmp(args.ar_locktable, sdp->sd_args.ar_locktable) || 1072 strcmp(args.ar_hostdata, sdp->sd_args.ar_hostdata)) 1073 return -EINVAL; 1074 1075 /* Some flags must not be changed */ 1076 if (args_neq(&args, &sdp->sd_args, spectator) || 1077 args_neq(&args, &sdp->sd_args, ignore_local_fs) || 1078 args_neq(&args, &sdp->sd_args, localflocks) || 1079 args_neq(&args, &sdp->sd_args, localcaching) || 1080 args_neq(&args, &sdp->sd_args, meta)) 1081 return -EINVAL; 1082 1083 if (sdp->sd_args.ar_spectator) 1084 *flags |= MS_RDONLY; 1085 1086 if ((sb->s_flags ^ *flags) & MS_RDONLY) { 1087 if (*flags & MS_RDONLY) 1088 error = gfs2_make_fs_ro(sdp); 1089 else 1090 error = gfs2_make_fs_rw(sdp); 1091 if (error) 1092 return error; 1093 } 1094 1095 sdp->sd_args = args; 1096 if (sdp->sd_args.ar_posix_acl) 1097 sb->s_flags |= MS_POSIXACL; 1098 else 1099 sb->s_flags &= ~MS_POSIXACL; 1100 spin_lock(>->gt_spin); 1101 gt->gt_log_flush_secs = args.ar_commit; 1102 spin_unlock(>->gt_spin); 1103 1104 gfs2_online_uevent(sdp); 1105 return 0; 1106 } 1107 1108 /** 1109 * gfs2_drop_inode - Drop an inode (test for remote unlink) 1110 * @inode: The inode to drop 1111 * 1112 * If we've received a callback on an iopen lock then its because a 1113 * remote node tried to deallocate the inode but failed due to this node 1114 * still having the inode open. Here we mark the link count zero 1115 * since we know that it must have reached zero if the GLF_DEMOTE flag 1116 * is set on the iopen glock. If we didn't do a disk read since the 1117 * remote node removed the final link then we might otherwise miss 1118 * this event. This check ensures that this node will deallocate the 1119 * inode's blocks, or alternatively pass the baton on to another 1120 * node for later deallocation. 1121 */ 1122 1123 static void gfs2_drop_inode(struct inode *inode) 1124 { 1125 struct gfs2_inode *ip = GFS2_I(inode); 1126 1127 if (test_bit(GIF_USER, &ip->i_flags) && inode->i_nlink) { 1128 struct gfs2_glock *gl = ip->i_iopen_gh.gh_gl; 1129 if (gl && test_bit(GLF_DEMOTE, &gl->gl_flags)) 1130 clear_nlink(inode); 1131 } 1132 generic_drop_inode(inode); 1133 } 1134 1135 /** 1136 * gfs2_clear_inode - Deallocate an inode when VFS is done with it 1137 * @inode: The VFS inode 1138 * 1139 */ 1140 1141 static void gfs2_clear_inode(struct inode *inode) 1142 { 1143 struct gfs2_inode *ip = GFS2_I(inode); 1144 1145 /* This tells us its a "real" inode and not one which only 1146 * serves to contain an address space (see rgrp.c, meta_io.c) 1147 * which therefore doesn't have its own glocks. 1148 */ 1149 if (test_bit(GIF_USER, &ip->i_flags)) { 1150 ip->i_gl->gl_object = NULL; 1151 gfs2_glock_put(ip->i_gl); 1152 ip->i_gl = NULL; 1153 if (ip->i_iopen_gh.gh_gl) { 1154 ip->i_iopen_gh.gh_gl->gl_object = NULL; 1155 gfs2_glock_dq_uninit(&ip->i_iopen_gh); 1156 } 1157 } 1158 } 1159 1160 static int is_ancestor(const struct dentry *d1, const struct dentry *d2) 1161 { 1162 do { 1163 if (d1 == d2) 1164 return 1; 1165 d1 = d1->d_parent; 1166 } while (!IS_ROOT(d1)); 1167 return 0; 1168 } 1169 1170 /** 1171 * gfs2_show_options - Show mount options for /proc/mounts 1172 * @s: seq_file structure 1173 * @mnt: vfsmount 1174 * 1175 * Returns: 0 on success or error code 1176 */ 1177 1178 static int gfs2_show_options(struct seq_file *s, struct vfsmount *mnt) 1179 { 1180 struct gfs2_sbd *sdp = mnt->mnt_sb->s_fs_info; 1181 struct gfs2_args *args = &sdp->sd_args; 1182 int lfsecs; 1183 1184 if (is_ancestor(mnt->mnt_root, sdp->sd_master_dir)) 1185 seq_printf(s, ",meta"); 1186 if (args->ar_lockproto[0]) 1187 seq_printf(s, ",lockproto=%s", args->ar_lockproto); 1188 if (args->ar_locktable[0]) 1189 seq_printf(s, ",locktable=%s", args->ar_locktable); 1190 if (args->ar_hostdata[0]) 1191 seq_printf(s, ",hostdata=%s", args->ar_hostdata); 1192 if (args->ar_spectator) 1193 seq_printf(s, ",spectator"); 1194 if (args->ar_ignore_local_fs) 1195 seq_printf(s, ",ignore_local_fs"); 1196 if (args->ar_localflocks) 1197 seq_printf(s, ",localflocks"); 1198 if (args->ar_localcaching) 1199 seq_printf(s, ",localcaching"); 1200 if (args->ar_debug) 1201 seq_printf(s, ",debug"); 1202 if (args->ar_upgrade) 1203 seq_printf(s, ",upgrade"); 1204 if (args->ar_posix_acl) 1205 seq_printf(s, ",acl"); 1206 if (args->ar_quota != GFS2_QUOTA_DEFAULT) { 1207 char *state; 1208 switch (args->ar_quota) { 1209 case GFS2_QUOTA_OFF: 1210 state = "off"; 1211 break; 1212 case GFS2_QUOTA_ACCOUNT: 1213 state = "account"; 1214 break; 1215 case GFS2_QUOTA_ON: 1216 state = "on"; 1217 break; 1218 default: 1219 state = "unknown"; 1220 break; 1221 } 1222 seq_printf(s, ",quota=%s", state); 1223 } 1224 if (args->ar_suiddir) 1225 seq_printf(s, ",suiddir"); 1226 if (args->ar_data != GFS2_DATA_DEFAULT) { 1227 char *state; 1228 switch (args->ar_data) { 1229 case GFS2_DATA_WRITEBACK: 1230 state = "writeback"; 1231 break; 1232 case GFS2_DATA_ORDERED: 1233 state = "ordered"; 1234 break; 1235 default: 1236 state = "unknown"; 1237 break; 1238 } 1239 seq_printf(s, ",data=%s", state); 1240 } 1241 if (args->ar_discard) 1242 seq_printf(s, ",discard"); 1243 lfsecs = sdp->sd_tune.gt_log_flush_secs; 1244 if (lfsecs != 60) 1245 seq_printf(s, ",commit=%d", lfsecs); 1246 if (args->ar_errors != GFS2_ERRORS_DEFAULT) { 1247 const char *state; 1248 1249 switch (args->ar_errors) { 1250 case GFS2_ERRORS_WITHDRAW: 1251 state = "withdraw"; 1252 break; 1253 case GFS2_ERRORS_PANIC: 1254 state = "panic"; 1255 break; 1256 default: 1257 state = "unknown"; 1258 break; 1259 } 1260 seq_printf(s, ",errors=%s", state); 1261 } 1262 return 0; 1263 } 1264 1265 /* 1266 * We have to (at the moment) hold the inodes main lock to cover 1267 * the gap between unlocking the shared lock on the iopen lock and 1268 * taking the exclusive lock. I'd rather do a shared -> exclusive 1269 * conversion on the iopen lock, but we can change that later. This 1270 * is safe, just less efficient. 1271 */ 1272 1273 static void gfs2_delete_inode(struct inode *inode) 1274 { 1275 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; 1276 struct gfs2_inode *ip = GFS2_I(inode); 1277 struct gfs2_holder gh; 1278 int error; 1279 1280 if (!test_bit(GIF_USER, &ip->i_flags)) 1281 goto out; 1282 1283 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &gh); 1284 if (unlikely(error)) { 1285 gfs2_glock_dq_uninit(&ip->i_iopen_gh); 1286 goto out; 1287 } 1288 1289 error = gfs2_check_blk_type(sdp, ip->i_no_addr, GFS2_BLKST_UNLINKED); 1290 if (error) 1291 goto out_truncate; 1292 1293 gfs2_glock_dq_wait(&ip->i_iopen_gh); 1294 gfs2_holder_reinit(LM_ST_EXCLUSIVE, LM_FLAG_TRY_1CB | GL_NOCACHE, &ip->i_iopen_gh); 1295 error = gfs2_glock_nq(&ip->i_iopen_gh); 1296 if (error) 1297 goto out_truncate; 1298 1299 if (S_ISDIR(inode->i_mode) && 1300 (ip->i_diskflags & GFS2_DIF_EXHASH)) { 1301 error = gfs2_dir_exhash_dealloc(ip); 1302 if (error) 1303 goto out_unlock; 1304 } 1305 1306 if (ip->i_eattr) { 1307 error = gfs2_ea_dealloc(ip); 1308 if (error) 1309 goto out_unlock; 1310 } 1311 1312 if (!gfs2_is_stuffed(ip)) { 1313 error = gfs2_file_dealloc(ip); 1314 if (error) 1315 goto out_unlock; 1316 } 1317 1318 error = gfs2_dinode_dealloc(ip); 1319 if (error) 1320 goto out_unlock; 1321 1322 out_truncate: 1323 error = gfs2_trans_begin(sdp, 0, sdp->sd_jdesc->jd_blocks); 1324 if (error) 1325 goto out_unlock; 1326 /* Needs to be done before glock release & also in a transaction */ 1327 truncate_inode_pages(&inode->i_data, 0); 1328 gfs2_trans_end(sdp); 1329 1330 out_unlock: 1331 if (test_bit(HIF_HOLDER, &ip->i_iopen_gh.gh_iflags)) 1332 gfs2_glock_dq(&ip->i_iopen_gh); 1333 gfs2_holder_uninit(&ip->i_iopen_gh); 1334 gfs2_glock_dq_uninit(&gh); 1335 if (error && error != GLR_TRYFAILED && error != -EROFS) 1336 fs_warn(sdp, "gfs2_delete_inode: %d\n", error); 1337 out: 1338 truncate_inode_pages(&inode->i_data, 0); 1339 clear_inode(inode); 1340 } 1341 1342 static struct inode *gfs2_alloc_inode(struct super_block *sb) 1343 { 1344 struct gfs2_inode *ip; 1345 1346 ip = kmem_cache_alloc(gfs2_inode_cachep, GFP_KERNEL); 1347 if (ip) { 1348 ip->i_flags = 0; 1349 ip->i_gl = NULL; 1350 } 1351 return &ip->i_inode; 1352 } 1353 1354 static void gfs2_destroy_inode(struct inode *inode) 1355 { 1356 kmem_cache_free(gfs2_inode_cachep, inode); 1357 } 1358 1359 const struct super_operations gfs2_super_ops = { 1360 .alloc_inode = gfs2_alloc_inode, 1361 .destroy_inode = gfs2_destroy_inode, 1362 .write_inode = gfs2_write_inode, 1363 .delete_inode = gfs2_delete_inode, 1364 .put_super = gfs2_put_super, 1365 .sync_fs = gfs2_sync_fs, 1366 .freeze_fs = gfs2_freeze, 1367 .unfreeze_fs = gfs2_unfreeze, 1368 .statfs = gfs2_statfs, 1369 .remount_fs = gfs2_remount_fs, 1370 .clear_inode = gfs2_clear_inode, 1371 .drop_inode = gfs2_drop_inode, 1372 .show_options = gfs2_show_options, 1373 }; 1374 1375